Extending the Foundational Model of Anatomy with Automatically Acquired Spatial Relations

Manuel Möller,Pinar Wennerberg,Michael Sintek,Sascha Seifert,Christian Folz

doi:10.1038/npre.2009.3471.1

Abstract

AbstractFormal ontologies have made significant impact in bioscience over the last ten years. Among them, the Foundational Model of Anatomy Ontology (FMA) is the most comprehensive model for the spatio-structural representation of human anatomy. In the research project MEDICO we use the FMA as our main source of background knowledge about human anatomy. Our ultimate goals are to use spatial knowledge from the FMA (1) to improve automatic parsing algorithms for 3D volume data sets generated by Computed Tomography and Magnetic Resonance Imaging and (2) to generate semantic annotations using the concepts from the FMA to allow semantic search on medical image repositories. We argue that in this context more spatial relation instances are needed than those currently available in the FMA. In this publication we present a technique for the automatic inductive acquisition of spatial relation instances by generalizing from expert-annotated volume datasets.

Highlights

Semantic medical image search as approached by Advances in medical imaging have greatly increased the amount of images produced in clinical facilities
We take low numbers for low evidence for this relation and do not take it into the model. minConfidence determines which fraction of each pair of source and destination concepts has to share the same direction before the pair and its predominant direction are added to the inferred model
If more than minConfidence of all tuples belong to the same class, we still add a representative of this class, taking their distribution as support for their universality

Summary

Introduction

Semantic medical image search as approached by Advances in medical imaging have greatly increased the amount of images produced in clinical facilities. Today's clinical facilities typically contain hospital information systems (HIS) for storing patient billing and accounting information, radiological information systems (RIS) for storing radiological reports, and picture archiving and control systems (PACS) for archiving medical images. It has become challenging for clinicians to query and retrieve relevant previous patient data due to the volume of information, the complexity and heterogeneous nature of today's information systems. The goal is to allow cross-lingual and modalityindependent search and retrieval across medical images, clinical findings and reports. This requires processes for automatic annotation of images and documents with concepts from formal ontologies to allow retrieval to be performed on an abstract level. Searching becomes independent of the concrete data representation and can leverage on the information modeled in formal ontologies, e.g., for query expansion as described in a recent ESWC publication[2]

Objectives

Results

Conclusion